ITBench-AA · Open Analysis

#	Model	Effort	Score	Cost / task	Harness
1	GPT-5.6 Sol	max	56.2%	–	–
2	Step 5 Preview	–	55.6%	–	–
3	Gemini 3.8 Flash	high	52.5%	–	–
4	GLM 5.3 Flash	max	51.2%	–	–
5	GPT-5.6 Terra	max	51.0%	–	–
6	Fable 5.1	max	49.5%	–	–
7	GPT-6 Sol	max	49.4%	–	–
8	GPT-6 Astra	max	48.6%	–	–
9	Kimi K3	max	47.7%	–	–
10	DeepSeek V4.1 Flash	max	46.9%	–	–
11	Opus 4.7	max	46.7%	–	–
12	GLM 5.3	max	46.1%	–	–
13	GPT-5.5	xhigh	45.8%	–	–
14	GLM 5.2	max	42.7%	–	–
15	Qwen 3.7 Max	max	42.5%	–	–
16	Grok 4.7	xhigh	42.1%	–	–
17	Gemini 3.5 Flash	high	40.3%	–	–
18	GPT-5.6 Luna	max	40.3%	–	–
19	Qwen 3.8 Max	max	40.3%	–	–
20	GLM 5.1	–	40.3%	–	–
21	Sonnet 4.6	max	39.8%	–	–
22	DeepSeek V4 Pro 0424	max	38.3%	–	–
23	Opus 5.5	max	38.2%	–	–
24	MiMo V2.5 Pro	–	38.2%	–	–
25	Gemma 4 31B	–	37.3%	–	–
26	Qwen 3.5 27B	–	35.5%	–	–
27	GPT-5.4 Mini	xhigh	35.2%	–	–
28	Qwen 3.5 397B-A17B	–	34.1%	–	–
29	Muse Spark 1.3	max	33.2%	–	–
30	Grok 4.3	high	32.7%	–	–
31	DeepSeek V4 Flash 0420	max	31.5%	–	–
32	Kimi K2.6	–	31.2%	–	–
33	Gemini 3.1 Pro	–	30.3%	–	–
34	Step 3.7 Flash	high	30.3%	–	–
35	Haiku 4.5	–	27.3%	–	–
36	MiniMax M2.7	–	26.5%	–	–
37	GPT-5.4 Nano	xhigh	24.4%	–	–
38	Gemma 4 26B	–	23.6%	–	–
39	Qwen 3.5 35B-A3B	–	21.5%	–	–
40	Grok 4.1 Fast	none	17.9%	–	–
41	gpt-oss-120b	high	5.6%	–	–
42	Nemotron 3 Super 120B A12B (Reasoning)	–	1.1%	–	–
43	Llama 3.3 70b	none	0.6%	–	–